H5-DataLoader

A data loader for using H5Dataset with PyTorch

Dependency

• Library
  • libpcl-dev
  • libopencv-dev
  • python3-numpy
• Python3
  • setuptools
  • wheel
  • scikit-build
  • cmake
  • ninja
  • numpy
  • h5py
  • opencv-python
  • pointsmap

Install

pip3 install git+https://github.com/shikishima-TasakiLab/h5dataloader

How to Use

Config

See H5-DataLoader-Config.

PyTorch

import torch
from torch.utils.data.dataloader import DataLoader
from h5dataloader.pytorch import HDF5Dataset

device = torch.device('cuda:0' if args.cuda else 'cpu')

train_dataset = HDF5Dataset(h5paths=['train1.hdf5', 'train2.hdf5', ...], config='config.json')
train_dataloader = DataLoader(h5dataset, batch_size=2, shuffle=True)

for train_itr, train_batch in enumerate(train_dataloader):
    img_rgb = train_batch['rgb'].to(device)
    img_gt = train_batch['gt'].to(device)
    ...